Tutorial: Getting Best Results

How to use the three commands effectively.

Prerequisites

• Claude Commands Library v5.0 installed
• Basic familiarity with Claude Code
• A project to work with

Part 1: Understanding Phase 0

Every command begins with Phase 0 — a shared validation layer that runs before any output.

Step 1: Analysis
  Detect language (English, Slovak, etc.)
  Identify prompt type (Task, Bug Fix, Question, etc.)
  Extract core intent

Step 2: Memory Recall
  Load known facts from project-profile.md
  Pre-fill tech stack, conventions, recent work

Step 3: Completeness Check
  Goal — what do you want to achieve?
  Context — project, technology, environment
  Scope — which files, components, areas
  Requirements — specific needs
  Constraints — limitations (optional)
  Expected Result — how to verify success

Step 4: Clarifying Questions
  Ask only what cannot be answered from memory
  Present options where multiple approaches exist

Step 5: Structured Output
  Goal / Context / Scope / Requirements / Constraints / Expected Result

Step 6: Approval Gate
  Wait for explicit approval before proceeding

The approval gate is mandatory — Claude never auto-executes after Phase 0.

Part 2: Using /prompt Effectively

The difference it makes

Before — vague prompt

/prompt Fix my code

Problems: no context, no scope, no expected result.

After — complete prompt

/prompt Fix the NullReferenceException in UserService.GetUser()
when the user does not exist. Should return null instead of throwing.
Using .NET 8, Entity Framework Core.

Why it works: clear goal, specific scope, stated expected behaviour, tech stack provided.

The 6 criteria

Every prompt should answer:

1. Goal — what do you want to achieve?
2. Context — project, technology, environment
3. Scope — which files, components, areas
4. Requirements — specific needs
5. Constraints — limitations, rules (optional)
6. Expected Result — how to verify success

Examples by task type

Bug fix:

/prompt Fix the authentication timeout in the React app.
Users are logged out after 5 minutes, expected 30 minutes.
Files: src/auth/AuthContext.tsx, src/api/client.ts
Stack: React 18, Axios, JWT

New feature:

/prompt Add dark mode toggle to the Settings page.
Should persist in localStorage and respect system preference on first load.
Stack: React, Tailwind CSS, shadcn/ui

Refactor:

/prompt Refactor UserService to use repository pattern.
Currently: direct DbContext calls in service layer.
Target: IUserRepository interface + concrete implementation.
Stack: .NET 8, EF Core

When Phase 0 asks questions

Answer fully — each question prevents misimplementation. If Claude asks which auth method, saying "JWT" saves a full refactor.

Part 3: Using /prompt-research Effectively

Give a focused goal

Too broad produces shallow results:

/prompt-research Understand the codebase

Focused produces deep results:

/prompt-research Understand the payment processing flow
and identify potential race conditions in order creation

What to expect

1. Claude spawns 2-5 agents in parallel (Explore, Pattern, Security, Performance, Citation)
2. First iteration maps the landscape — files, entry points, key paths
3. Subsequent iterations (2-4 total) drill into gaps from the previous round
4. Final report: structured findings with file:line citations

Good research prompts

/prompt-research Map the authentication system and
identify any JWT validation gaps

/prompt-research Understand how database migrations work,
what the schema history is, and flag any missing rollbacks

/prompt-research Find all API endpoints that handle payment data
and check for OWASP Top 10 issues

Using the output

The report cites every claim with a file path and line number. Use these to:

• Jump directly to the relevant code
• Verify findings before acting on them
• Create precise follow-up prompts with /prompt

Part 4: Using /prompt-article-readme Effectively

Run it from the project root

cd /my-project
/prompt-article-readme

Claude scans the directory it's running in. Running from a subdirectory produces a scoped README, not a top-level one.

Choosing a style level

When prompted, choose based on audience:

• Minimal — personal projects, internal tools
• Standard — open source, team projects
• Comprehensive — public libraries, portfolio pieces

Updating an existing README

If a README already exists, Claude reads it first and proposes targeted updates — it does not overwrite everything. Review the diff before approving.

Part 5: Iterating on Output

Feedback responses

After any command output, you can respond:

Response	What happens
y / yes	Approve and proceed
partial	Output was mostly right, needs adjustment — Claude asks what to fix
wrong	Output missed the mark — Claude asks what specifically was wrong
explain	Claude explains its reasoning before you decide
options	Claude presents alternative approaches

Refining a result

User: partial

Claude: What needs adjustment?

User: The session duration logic is right, but you modified
AuthContext.tsx — the actual timeout is handled in api/client.ts

Claude: [corrects the target file]

Part 6: Common Mistakes

Too vague:

# Wrong
/prompt Fix the bug

# Right
/prompt Fix the login button not responding on mobile Safari.
React 18, Tailwind CSS.

Not providing tech stack: Claude has to guess or ask. Include Stack: ... in any prompt that touches code.

Accepting 80% output: Use partial and refine. Getting to 100% takes less time than fixing broken code.

Over-trusting output: Always review generated code, test before deploying, verify facts. You remain responsible.

Too broad a research goal: /prompt-research with "understand everything" spreads agents thin. Narrow the scope.

Part 7: Pro Tips

Meta-prompting — if unsure how to phrase something:

/prompt Help me write a prompt for implementing
OAuth2 authentication in a .NET 8 minimal API

Use examples — show the expected input/output shape:

/prompt Create a validation function.
Input: { name: "", age: -1 }
Output: { valid: false, errors: ["name required", "age must be positive"] }

Chain commands — use /prompt to structure a task, then hand the output to Claude for implementation. Use /prompt-research first when you don't know the codebase well.

Specify output format:

/prompt Explain the Observer pattern.
Output as: 1-sentence summary, when to use (3 bullets), TypeScript example, pitfalls.

---

Related

• AI Fluency Framework
• Phase 0 Flow
• Multi-Agent Research
• Commands Reference